WO2022186607A1

WO2022186607A1 - Method for acquiring urination information with high accuracy

Info

Publication number: WO2022186607A1
Application number: PCT/KR2022/002956
Authority: WO
Inventors: 송지영; 두경연; 정지영; 김대연
Original assignee: 다인기술 주식회사
Priority date: 2021-03-02
Filing date: 2022-03-02
Publication date: 2022-09-09
Also published as: US11763802B2; US20230377566A1; US20220392438A1

Abstract

According to one embodiment of the present specification, provided is a method for acquiring urination information with high accuracy, the method comprising: dividing acoustic data into a plurality of windows; acquiring, from the acoustic data, fragment target data corresponding to each window; using the acquired fragment target data to acquire fragment uroflow data and fragment classification data for dividing a urination section or a non-urination section; and acquiring urination data by using the acquired fragment classification data and fragment uroflow data.

Description

How to obtain high-accuracy urination information

The present specification relates to a method of obtaining urination information with high accuracy, and more particularly, to a method of extracting urination information from sound data related to urination recorded using a urination/non-urination section classification model and a urine velocity prediction model will be.

Sound generated from a person's body has been used as important information in confirming the person's health status or disease. In particular, since the sound generated during the urination process of a person may include important information for diagnosing the person's urination function, research on a method of obtaining urination information by analyzing the urination sound is continuously being conducted. Specifically, the urination rate or amount of urination of the person may be predicted by analyzing the acoustic data obtained by recording the urination process of the person.

Meanwhile, the sound generated during the urination process of a person may include various sounds generated by the surrounding environment in addition to the sound related to urination. The sound caused by the surrounding environment may be included in the urination section or the non-urination section during the urination process. there was.

Therefore, in order to obtain more accurate urination information, a method of analyzing sound data for a urination process is required in consideration of a case in which a sound from the surrounding environment is included in a non-urination section.

An object to be solved in the present specification is to provide a method for obtaining urination information for obtaining urination information with high accuracy from acoustic data obtained by recording a human urination process.

An object to be solved in the present specification is to provide a method for acquiring urination information using a model for dividing a urination process into a urination section and a non-urination section and a model for predicting a urine velocity in the urination process.

An object to be solved in the present specification is to provide a method of first determining whether a urination process exists with respect to the collected acoustic data and then performing a urine velocity prediction.

The problem to be solved in this specification is not limited to the above-mentioned problems, and the problems not mentioned will be clearly understood by those of ordinary skill in the art to which the present invention belongs from the present specification and the accompanying drawings. .

According to an embodiment of the present specification, there is provided a method for obtaining urination information with high accuracy, the method comprising: obtaining sound data, the sound data having a start point and an end point; Acquiring first to mth engraving target data corresponding to m windows from the sound data - Each of the m windows has a predetermined time interval, and continuously between the start point and the end point is determined and consecutive windows among the m windows partially overlap each other, and m is a natural number equal to or greater than 2; Inputting the first to mth fragment target data into a pre-trained urination/non-urination classification model to obtain first to mth fragment classification data - The urination/non-urination classification model forms a first data matrix learning to receive input and output data including at least one value for classifying a urination section or a non-urination section; Acquiring first to n-th engraving target data corresponding to n windows from the sound data - The n windows each have a predetermined time interval, and are continuously determined between the start point and the end point. consecutive windows among the n windows partially overlap each other, and n is a natural number of 2 or more and m or less; Obtaining the first to n-th piece yaw velocity data by inputting the first to n-th piece target data into a pre-trained yaw velocity prediction model - The yaw velocity prediction model receives a second data matrix and at least a value related to the yaw velocity learned to output data containing one or more; and obtaining urination data using at least the first to m-th piece classification data and the first to n-th piece urine velocity data.

According to another embodiment of the present specification, in a method of obtaining urination information with high accuracy, the step of acquiring sound data by sampling an acoustic signal obtained by recording a urination process through an external device -The above acoustic data has a start point and an end point; Acquiring first to n-th engraving target data corresponding to first to n-th windows from the sound data - The first to n-th windows each have a predetermined time interval, the starting point and the determined continuously between the end points, wherein n is a natural number greater than or equal to 2; Inputting at least a portion of the first to n-th piece target data into a pre-trained urination/non-urination classification model to obtain first to m-th piece classification data - The urination/non-urination classification model is a first receiving a data matrix and learning to output data including at least one value for classifying a urination section or a non-urination section, wherein m is a natural number greater than or equal to 2 and less than or equal to n; Obtaining the first to n-th piece yaw velocity data by inputting the first to n-th piece target data into a pre-trained yaw velocity prediction model - The yaw velocity prediction model receives a second data matrix and at least a value related to the yaw velocity learned to output data containing one or more; and obtaining urination data using at least the first to m-th piece classification data and the first to n-th piece urine velocity data.

Solutions of the problems of the present specification are not limited to the above-described solutions, and solutions not mentioned will be clearly understood by those of ordinary skill in the art to which the present invention belongs from the present specification and the accompanying drawings. will be able

According to an example of the present specification, by using the result of classifying the urination process into a urination section and a non-urination section and a result of predicting the urine speed in the urination process together, it is possible to obtain a highly accurate urine velocity prediction result for the urination process.

According to an example of the present specification, by using data obtained by classifying the urination process into a urination section and a non-urination section, it is possible to prevent the urine rate data in the non-urination section from being included in the urine velocity prediction result of the urination process.

According to an example of the present specification, it is possible to prevent unnecessary data analysis in advance by first determining the presence or absence of a urination process prior to predicting the urinary velocity.

Effects according to the present specification are not limited to the above-described effects, and effects not mentioned will be clearly understood by those of ordinary skill in the art to which the present invention pertains from this specification and the accompanying drawings.

1 is a diagram illustrating an environment for analyzing urination information according to an embodiment of the present specification.

2 is a diagram illustrating a configuration of an acoustic analysis system according to an embodiment of the present specification.

3 is a diagram illustrating a process in which a configuration of an acoustic analysis system according to an embodiment of the present specification operates.

4 and 5 are diagrams illustrating a method of classifying sound data according to a window according to an embodiment of the present specification.

6 is a diagram illustrating a process of extracting a feature value from sound data according to an embodiment of the present specification.

7 and 8 are diagrams illustrating a process of acquiring target data to be analyzed according to an embodiment of the present specification.

9 is a diagram illustrating a process of acquiring yaw velocity data using a yaw velocity prediction model according to an embodiment of the present specification.

10 is a diagram illustrating a method of obtaining candidate yaw velocity data according to an embodiment of the present specification.

11 is a diagram illustrating a process of acquiring classification data using a urination/non-urination classification model according to an embodiment of the present specification.

12 is a diagram illustrating a method of acquiring urination classification data according to an embodiment of the present specification.

13 is a diagram illustrating a method of acquiring urination classification data according to another embodiment of the present specification.

14 and 15 are diagrams illustrating a method of acquiring urination data using candidate urine velocity data and voiding classification data according to an embodiment of the present specification.

16 and 17 are flowcharts illustrating a method for analyzing urination information according to an embodiment of the present specification.

18 is a diagram illustrating a graph comparing the results of not using and using the urination/non-urination classification model according to an embodiment of the present specification.

19 is a flowchart illustrating a process of learning a yaw velocity prediction model according to an embodiment of the present specification.

20 is a flowchart illustrating a process of training a urination/non-urination classification model according to an embodiment of the present specification.

21 is a diagram illustrating measured data and predicted data according to an embodiment of the present specification.

22 is a diagram illustrating actual measurement data, prediction data, and corrected prediction data according to an embodiment of the present specification.

23 is a flowchart illustrating a data correction method according to an embodiment of the present specification.

24 is a diagram illustrating a data correction process according to an embodiment of the present specification.

25 is a graph illustrating a relationship between measured data and predicted data before data correction according to an embodiment of the present specification.

26 is a graph illustrating a relationship between measured data and predicted data after data correction according to an embodiment of the present specification.

An overlapping degree of successive windows among the m windows may be different from an overlapping degree of successive windows among the n windows.

An overlapping degree of successive windows among the m windows may be smaller than an overlapping degree of successive windows among the n windows.

The first to mth engraving target data and the first to nth engraving target data may each include a feature value extracted from at least a portion of the sound data.

The acquiring of the first to mth pieces of target data may include: converting the sound data into spectrogram data; and obtaining the first to mth slice target data corresponding to the first to mth windows from the spectogram data.

The acquiring of the first to mth engraving target data may include: acquiring first to mth sculptural sound data corresponding to the first to mth windows; and converting the first to mth sculptural sound data into spectrogram data to obtain the first to mth sculptural target data.

The spectrogram data may be Mel-spectrogram data to which a Mel-filter is applied.

The first to m-th piece classification data is obtained by sequentially inputting the first to m-th piece target data to the urination/non-urination classification model, and the first to n-th piece urine velocity data is the urine velocity prediction model It may be obtained by sequentially inputting the first to n-th piece target data into the .

The urination/non-urination classification model comprises at least a first input layer, a first convolution layer, a first hidden layer, and a first output layer, and , the yaw velocity prediction model may include at least a second input layer, a second convolution layer, a second hidden layer, and a second output layer.

A size of the first data matrix may be the same as a size of the second data matrix.

The acquiring of the urination data may include: acquiring urination classification data using the first to m-th piece classification data; obtaining candidate yaw velocity data using the first to nth piece yaw velocity data; and obtaining the urination data by processing the candidate urine velocity data using the urination classification data.

The urination data may be obtained by performing a convolution operation on the urination classification data and the candidate urine velocity data.

The method for obtaining the urination information includes: determining whether or not a urination process is present with respect to the sound data using the first to mth fragment classification data obtained after the acquiring of the first to mth fragment classification data is performed further comprising; if it is determined that a urination process exists with respect to the sound data, obtaining the first to nth pieces of target data from the sound data; Obtaining the first to nth piece urine velocity data by input to the model, and obtaining the urination data using at least the first to mth piece classification data and the first to nth piece urine velocity data can be performed.

According to another embodiment of the present specification, in a method for obtaining urination information with high accuracy, the step of obtaining sound data by sampling an acoustic signal obtained by recording a urination process through an external device - The sound data includes a starting point and having an end point-; Acquiring first to n-th engraving target data corresponding to the first to n-th windows from the sound data - The first to n-th windows each have a predetermined time interval, and between the start point and the end point is determined continuously in , wherein n is a natural number greater than or equal to 2; Inputting at least a portion of the first to n-th piece target data into a pre-trained urination/non-urination classification model to obtain first to m-th piece classification data - The urination/non-urination classification model is a first receiving a data matrix and learning to output data including at least one value for classifying a urination section or a non-urination section, wherein m is a natural number greater than or equal to 2 and less than or equal to n; Obtaining the first to n-th piece yaw velocity data by inputting the first to n-th piece target data into a pre-trained yaw velocity prediction model - The yaw velocity prediction model receives a second data matrix and at least a value related to the yaw velocity learned to output data containing one or more; and obtaining urination data using at least the first to m-th piece classification data and the first to n-th piece urine velocity data.

Two consecutive windows among the first to nth windows may overlap.

Two consecutive windows among the first to nth windows do not overlap each other, and m may be the same as n.

Each of the first to nth pieces of target data may include a feature value extracted from at least a portion of the sound data.

The obtaining of the first to nth pieces of target data may include: converting the sound data into spectrogram data; and obtaining the first to nth slice target data corresponding to the first to nth windows from the spectogram data.

The acquiring of the first to n-th engraving target data may include: acquiring first to n-th engraving sound data corresponding to the first to n-th windows; and converting the first to nth sculptural sound data into spectrogram data, respectively, to obtain the first to nth engraving target data.

The spectrogram data may be Mel-spectrogram data.

The first to mth piece classification data are obtained by sequentially inputting m pieces of piece target data corresponding to m non-overlapping windows among the first to nth windows to the urination/non-urination classification model, The first to nth piece yaw velocity data may be obtained by sequentially inputting the first to nth piece yaw velocity data into the yaw velocity prediction model.

The voiding/non-urination classification model includes at least a first input layer, a first convolutional layer, a first hidden layer, and a first output layer, and the urine velocity prediction model includes at least a second input layer, a second convolution It may include a layer, a second hidden layer, and a second output layer, and a data type input to the first input layer and a data type input to the second input layer may be the same.

According to another embodiment of the present specification, obtaining a measured data group for a plurality of entities in a data collection period, wherein the measured data group includes at least first measured data including a measurement value of a urination amount of the first entity and a second entity including second measured data including a measurement value of the amount of urination; obtaining a prediction data group for the urination process of the plurality of individuals in the data collection period, wherein the prediction data group includes first prediction data including a voiding amount predicted value of the first individual and a urination amount prediction of the second individual and second prediction data including a value, wherein the predicted urination amount of the first object is generated from first sound data recorded with the urination process of the first object, and the predicted urination amount of the second object is the second generated from second acoustic data recording the subject's urination process; obtaining a compensation value using the measured data group and the predicted data group; Obtaining the target prediction data for the urination process of the target object after the data collection period - The target prediction data includes a urination amount predicted value for the urination process of the target object, the urination amount predicted value of the target object is the target generated from subject acoustic data recording the subject's urination process; and correcting the target prediction data using the correction value.

The obtaining of the correction value may include: generating a data set group using the measured data group and the predicted data group; and calculating the correction value from the data set group using a regression analysis technique.

The data set group may include a first data set generated by matching a representative value of the urination amount measurement values included in the first actual data to a representative value of the urination amount predicted values included in the first prediction data.

The data set group is a representative value of the urine quantity measurement values of a first time interval among the urine quantity measurement values of the plurality of days included in the first measured data, among the urine quantity measurement values of the plurality of days included in the second actual measurement data. The first data set may include a first data set generated by matching representative values of measurement values of urination in a second time interval, and the length of the first time interval may be the same as the length of the second time interval.

The regression analysis technique may be a linear regression analysis technique, and the correction value may be a slope of a function obtained using the linear regression analysis technique.

In the step of calculating the correction value from the data set group by using the regression analysis technique, a value obtained from the first measured data is used as either an independent variable or a dependent variable, and obtained from the first prediction data The value is used as the other of the independent variable and the dependent variable, and the correction value may be a regression coefficient calculated through the regression analysis technique.

The acquiring of the target prediction data may include: generating target urine velocity data for a urination process of the target entity using the target acoustic data; and calculating a urination amount predicted value for the urination process of the target individual from the target urine velocity data.

The obtaining of the measured data group and the obtaining of the predicted data group may be performed simultaneously.

According to another embodiment of the present specification, the method may include: acquiring actual measurement data reflecting an amount of urination measured in a urination process of a target object in a first period, wherein the measured data includes at least one measurement value of the amount of urination; acquiring an acoustic data group in which the urination process of the target object is recorded in the first period, wherein the acoustic data group includes first acoustic data in which a first urination process is recorded; obtaining predictive data using the sound data group, wherein the predictive data includes a voiding amount predictive value calculated from the first sound data; obtaining a compensation value using the measured data and the predicted data; acquiring second sound data by recording a second urination process of the target object in a second period after the first period; calculating a target urination amount predicted value from the second sound data; and correcting the target urination amount predicted value by using the correction value.

In the obtaining of the correction value, a regression analysis technique is used, the value obtained from the measured data is used as either an independent variable or a dependent variable, and the value obtained from the prediction data is used as the other of the independent variable or the dependent variable. used and the correction value may be a regression coefficient calculated through the regression analysis technique.

The value obtained from the actual measurement data may be a urination amount measurement value corresponding to the first urination process during the first period, and the value obtained from the prediction data may be a urination amount predicted value calculated from the first sound data.

Calculating the target urination amount predicted value from the second acoustic data may include: generating target urine velocity data using the second acoustic data; and calculating the target urination amount predicted value from the target urine velocity data.

According to another embodiment of the present specification, the method may include: acquiring actually measured data reflecting the amount of urination measured in the urination process of the subject in a first period, wherein the measured data includes a plurality of measurement values of the amount of urination; acquiring sound data by recording the urination process of the target object in a second period after the first period; obtaining urination amount prediction data using the sound data; and correcting the urination amount prediction data by using the statistical values of the actually measured data, wherein the statistical values are the mode, median, average, variance, and A data correction method that is at least one of standard deviation values may be provided.

The above-mentioned objects, features and advantages of the present specification will become more apparent from the following detailed description in conjunction with the accompanying drawings. However, since the present specification may have various changes and may have various embodiments, specific embodiments will be exemplified in the drawings and described in detail below.

Throughout the specification, like reference numerals refer to like elements in principle. In addition, components having the same function within the scope of the same idea shown in the drawings of each embodiment will be described using the same reference numerals, and overlapping descriptions thereof will be omitted.

Numbers (eg, first, second, etc.) used in the description process of the present specification are only identifiers for distinguishing one component from other components.

In addition, the suffixes "module" and "part" for the components used in the following embodiments are given or mixed in consideration of only the ease of writing the specification, and do not have distinct meanings or roles by themselves.

Unless specifically stated or clear from the context, the term "about" in reference to a numerical value shall be understood to mean the stated numerical value and up to +/−10% of that value, and in reference to a numerical range "about" The term " can be understood to mean a range from a value 10% lower than the lower limit of the numerical range to a value 10% higher than the upper limit of the numerical range.

In the following examples, the singular expression includes the plural expression unless the context clearly dictates otherwise.

In the following examples, terms such as "comprise" or "have" mean that the feature or element described in the specification is present, and the possibility of adding one or more other features or elements is excluded in advance. it is not doing

In the drawings, the size of the components may be exaggerated or reduced for convenience of description. For example, the size and thickness of each component shown in the drawings are arbitrarily indicated for convenience of description, and the present invention is not necessarily limited to the illustrated bar.

In cases where certain embodiments are otherwise practicable, a specific process sequence may be performed different from the described sequence. For example, two processes described in succession may be performed substantially simultaneously, or may be performed in an order opposite to the order described.

In the following embodiments, when a film, region, or component is connected, other films, regions, and components are interposed between the films, regions, and components as well as when the films, regions, and components are directly connected. It also includes cases where it is indirectly connected.

For example, in the present specification, when it is said that a film, a region, a component, etc. are electrically connected, not only the case where the film, a region, a component, etc. are directly electrically connected, but also other films, regions, and components are interposed therebetween. Indirect electrical connection is also included.

The present specification relates to a method of obtaining urination information with high accuracy, and more particularly, a urination/non-urination classification model that records a human urination process to obtain acoustic data, and divides the urination section into a urination section and a non-urination section, and The present invention relates to a method of obtaining urination data by analyzing acoustic data using a urine velocity prediction model, and obtaining urination information using the obtained urination data.

Hereinafter, with reference to FIG. 1, a general environment in which the above-described method of obtaining urination information is performed will be described.

1 is a diagram illustrating an environment for analyzing urination information according to an embodiment of the present specification. Referring to FIG. 1 , an acoustic analysis system 1000 , a recording device 2000 , and an external server 3000 may be used to obtain urination information for a urination process.

The acoustic analysis system 1000 may acquire urination information based on data about the urination process. For example, the acoustic analysis system 1000 may acquire acoustic data recorded with a urination process from the recording apparatus 2000 , and acquire urination information using the acquired acoustic data. A process in which the acoustic analysis system 1000 acquires urination information from the acoustic data will be described in detail later.

The acoustic analysis system 1000 may communicate with the external server 3000 . The sound analysis system 1000 may obtain the above-described sound data from the recording apparatus 2000 or may also obtain it from the external server 3000 . Also, the sound analysis system 1000 may provide urination information obtained by analyzing the sound data to the external server 3000 . In other words, the acoustic analysis system 1000 may obtain urination information by analyzing acoustic data for a urination process received from the outside, and output or provide it to the outside.

The recording apparatus 2000 may record a urination-related sound. Specifically, the recording apparatus 2000 may be worn by a person or installed in a space where a person urinates to record a sound generated during a urination process. For example, the recording device 2000 is a wearable device equipped with a recording function, such as a smart watch, a smart band, a smart ring, and a smart neckless. It may include a smart phone, tablet, desktop, laptop, portable recorder, or installation recorder.

The recording apparatus 2000 may acquire sound data by recording a urination process.

Here, the sound data may be obtained by digitizing an analog sound signal for the urination process. For example, the recording device 2000 includes an Analog to Digital Converter (ADC) module, and a specific sampling rate such as 8 kHz, 16 kHz, 22 kHz, 32 kHz, 44.1 kHz, 48 kHz, 96 kHz, 192 kHz, or 384 kHz. can be used to obtain acoustic data from the acoustic signal for the urination process.

The recording apparatus 2000 may provide the acquired sound data to the sound analysis system 1000 and/or the external server 3000 . To this end, the recording apparatus 2000 may perform wired and/or wireless data communication with the acoustic analysis system 1000 and/or the external server 3000 .

The recording apparatus 2000 may also be used as a means for delivering urination information to the user. For example, the recording apparatus 2000 may obtain urination information from the sound analysis system 1000 and output it to the user.

The external server 3000 may store or provide various data. For example, the external server 3000 may store acoustic data acquired from the recording apparatus 2000 or urination information acquired from the acoustic analysis system 1000 . As another example, the external server 3000 provides the acoustic data obtained from the recording device 2000 to the acoustic analysis system 1000 , and provides the urination information obtained from the acoustic analysis system 1000 to the recording device 2000 . can provide

Meanwhile, the acoustic analysis system 1000 and the recording apparatus 2000 may be implemented as one device. For example, the sound analysis system 1000 may acquire sound data by including a module having a recording function by itself. As another example, the components of the sound analysis system 1000 may be built in the recording apparatus 2000 so that the recording apparatus 2000 may provide a function of analyzing sound data by itself.

음향 분석 시스템acoustic analysis system

Hereinafter, a configuration of the acoustic analysis system 1000 and an operation process thereof will be described in detail with reference to FIGS. 2 and 3 .

2 is a diagram illustrating a configuration of an acoustic analysis system 1000 according to an embodiment of the present specification.

3 is a diagram illustrating a process in which the configuration of the acoustic analysis system 1000 according to an embodiment of the present specification operates.

Referring to FIG. 2 , the acoustic analysis system 1000 includes a preprocessor 1100 , a feature extraction unit 1200 , a urine velocity prediction unit 1300 , a urination/non-urination classification unit 1400 , and a urination information extraction unit 1500 . ), an input unit 1600 , an output unit 1700 , a communication unit 1800 , and a control unit 1900 .

The preprocessor 1100 may pre-process the sound data received by the sound analysis system 1000 . Pre-processing is a process performed prior to extracting feature values from sound data, and may include filtering as described below.

The preprocessor 1100 may perform filtering for noise removal on the sound data. Here, the filtering may refer to a process of excluding noise-related data from sound data, and for this purpose, a high-pass filter, a low-pass filter, and a band-pass filter are used. etc. may be used. The filtering of the preprocessor 1100 may be omitted.

Meanwhile, in the preprocessor 1100 , windowing, which will be described later, may be performed.

The feature extraction unit 1200 may extract a feature value from the preprocessed sound data. Here, the feature value may mean digitizing the unique features of the sound data. For example, the feature value is a time-domain spectral magnitude value, spectral centroid, frequency-band spectral magnitude value, frequency-band root mean square (RMS) value, spectrogram magnitude value, Mel-spectrogram magnitude value, and Bispectrum. Score (BGS), Non-Gaussianity Score (NGS), Formants Frequencies (FF), Log Energy (LogE), Zero Crossing Rate (ZCR), Kurtosis (Kurt), and Mel-frequency cepstral coefficient (MFCC) The feature extractor 1200 may convert the preprocessed sound data and extract a feature value therefrom. The feature extraction unit 1200 may change the conversion form according to a feature value to be extracted from the sound data. For example, when the feature value to be extracted is a spectral value, the feature extractor 1200 may convert the preprocessed sound data into spectral data having a frequency axis. As another example, when the feature value to be extracted is a mel-spectrogram image value, the feature extractor 1200 may convert the preprocessed sound data into spectrogram image data having a time axis and a frequency axis. If there are a plurality of types of feature values to be extracted, the feature extraction unit 1200 may convert the preprocessed sound data into various types of data. Hereinafter, for convenience of explanation, a case where the feature values to be extracted by the feature extraction unit 1200 are values of a mel-spectrogram image is mainly described, but the technical spirit of the present specification is not limited thereto, and the feature values are different forms. It can be applied similarly to the case of .

The feature extractor 1200 may acquire engraving target data through windowing on Mel-spectrogram image data obtained by converting pre-processed sound data. Here, the windowing may refer to discriminating between the start point and the end point of sound data or converted data using a window having a time interval of a certain size. A detailed windowing method will be described later.

The fragment target data obtained through windowing may be understood as vector data in each window section, matrix data, or data having other formats. For example, the engraving target data may be understood as a set of vector data in which values of the above-described Mel-spectrogram image are arranged in a line in each window section. As another example, the engraving target data may be understood as a set of matrix-type data in which the values of the above-described mel-spectrogram image are taken into consideration in each window section on a time axis and a frequency axis.

Meanwhile, the feature extraction unit 1200 may divide the sound data into different pieces of sound data through windowing before converting the sound data and extracting the feature values. Alternatively, windowing is performed in the pre-processing unit 1100 , the feature extracting unit 1200 obtains the piece sound data from the pre-processing unit 1100 , and the obtained piece sound data is used as data including a feature value (eg, a spectrum). , spectrogram, or Mel-spectrogram) to extract a feature value, and data in a vector form, a matrix form, or other form of the extracted feature values may be obtained as engraving target data.

The feature extractor 1200 may transmit the piece target data to the urine velocity predictor 1300 and/or the urination/non-urination classification unit 1400 .

The urine velocity prediction unit 1300 may predict the urine velocity during urination. For example, the yaw velocity predictor 1300 may calculate predicted yaw velocity values for the acoustic data using a pre-learned yaw velocity prediction model.

The yaw velocity prediction model may refer to a model trained using machine learning. Here, machine learning may be understood as a comprehensive concept including an artificial neural network and further deep-learning. For example, the urine velocity prediction model may be implemented as an artificial neural network trained with a training data set in which the data obtained by precisely measuring the urination rate in the urination process are labeled with the acoustic data acquired by recording the urination process. The structure and learning method of the yaw velocity prediction model will be described in detail later.

The urine velocity prediction unit 1300 may obtain urine velocity values according to time in the urination process using the urine velocity prediction model and provide the urine velocity values to the urination information extractor 1500 .

The urination/non-urination classification unit 1400 may classify a urination section and a non-urination section in the urination process. For example, the urination/non-urination classification unit 1400 may obtain classification data obtained by dividing urination sections and non-urination sections from the acoustic data using a pre-trained urination/non-urination classification model.

The urination/non-urination classification model may refer to a model trained using machine learning. For example, in the urination/non-urination classification model, data indicating the urination/non-urination section obtained using precisely measured urination data during the urination process is labeled on the acoustic data obtained by recording the urination process. It can be implemented as an artificial neural network trained with a training data set. The structure and learning method of the urination/non-urination classification model will be described in detail later.

The urination/non-urination classification unit 1400 acquires classification values indicating whether urination/non-urination according to time in the urination process using the urination/non-urination classification model, and sends the classification values to the urination information extraction unit 1500 can provide

The urination information extractor 1500 may acquire urination information for a urination process. For example, the urination information extraction unit 1500 may generate urination data using the urine velocity values obtained from the above-described urine velocity prediction unit 1300 and the classification values obtained from the urination/non-urination classification unit 1400. and urination information may be extracted from the generated urination data.

Specifically, the urination information extractor 1500 may generate candidate urine velocity data for the recorded urination process by using the obtained urine velocity values. Also, the urination information extraction unit 1500 may generate urination classification data for the recorded urination process by using the obtained classification values. The urination information extractor 1500 may acquire urination data based on the candidate urine velocity data and the urination classification data described above by using the urination data calculation model. Here, the urination data may be understood as a set of urine velocity values according to time for a recorded urination process.

The urination information extraction unit 1500 may extract urination information from the urination data. Here, the voiding information includes a maximum flow rate, an average flow rate, a voiding amount, a urination start and end time, a flow time, and a time to maximum flow in the urination process. rate) and urination time (with or without interruption time), and the like.

The above-described preprocessor 1100 , feature extraction unit 1200 , urine velocity prediction unit 1300 , urination/non-urination classification unit 1400 , and urination information extraction unit 1500 may refer to software programs. For example, the noise removal and windowing process of the preprocessor 1100, the data conversion and target data generation process of the feature extraction unit 1200, the urine velocity prediction model of the urine velocity prediction unit 1300, urination / non-urination classification unit The urination/non-urination classification model of 1400 and the urination data calculation model of the urination information extraction unit 1500 are in the memory unit (not shown) of the acoustic analysis system 1000 to be described later in the form of a plurality of functions or commands. It may be stored and executed by being loaded by the controller 1900 .

The input unit 1600 may receive a user input from a user. The user input may be made in various forms, including a key input, a touch input, and a voice sound. Examples of the input unit 1600 include a traditional keypad, keyboard, and mouse, as well as a touch sensor for sensing a user's touch, and various types of input means for sensing or receiving various types of user input. Concept.

The output unit 1700 may output urination information and provide it to the user. The output unit 1700 is a comprehensive concept including a display that outputs an image, a speaker that outputs a sound, a haptic device that generates vibration, and other various types of output means.

The communication unit 1800 may communicate with an external device. The sound analysis system 1000 may transmit/receive data to and from the recording device 2000 or the external server 3000 through the communication unit 1800 . For example, the acoustic analysis system 1000 may provide urination information to the recording device 2000 and/or the external server 3000 through the communication unit 1800 , and transmit the acoustic data to the recording device 2000 and/or the external server 3000 . It can be received from the external server (3000).

The controller 1900 may control the overall operation of the acoustic analysis system 1000 . For example, the control unit 1900 may include the preprocessor 1100 , the feature extraction unit 1200 , the urine velocity prediction unit 1300 , the urination/non-urination classification unit 1400 , and the urination information extraction unit 1500 related to By loading and executing the program, it is possible to obtain urination information from the sound data. The controller 1900 may be implemented as a central processing unit (CPU) or a similar device according to hardware, software, or a combination thereof. In hardware, it may be provided in the form of an electronic circuit that performs a control function by processing an electrical signal, and in software, it may be provided in the form of a program or code for driving a hardware circuit.

The acoustic analysis system 1000 may further include a memory unit for storing various types of information. Various data may be temporarily or semi-permanently stored in the memory unit. Examples of the memory unit may include a hard disk (HDD), a solid state drive (SSD), a flash memory, a read-only memory (ROM), and a random access memory (RAM). have. The memory unit may be provided in a form embedded in the sound analysis system 1000 or in a form detachable.

음향 분석 과정Acoustic Analysis Course

Hereinafter, each step of the acoustic analysis process performed in the aforementioned acoustic analysis system 1000 will be described in detail with reference to FIGS. 4 to 14 .

윈도우 구분 방법How to separate windows

4 and 5 are diagrams illustrating a method of classifying sound data according to a window according to an embodiment of the present specification. Hereinafter, a method of dividing a window with respect to sound data will be described for convenience of explanation, but the window dividing method may be equally applied not only to sound data but also to data obtained by converting sound data to extract feature values.

The acoustic data may have a starting point and an ending point. In this case, the length of the sound data may be determined as the time between the start point and the end point. The length of the sound data may be determined according to the length at which the sound data is recorded.

The sound data may be divided into at least one window. For example, the sound data may be divided into at least one or more windows sequentially determined between the start point and the end point, and pieces of sound data corresponding to each window may be obtained. Here, the sequential determination of at least one window means that the plurality of windows are sequentially arranged from the start point to the end point of the sound data. Meanwhile, at least one window may be assigned a specific section between the start point and the end point of the sound data in time series. The window may have a predetermined size. The size of the window may be determined according to the length of the sound data and the number of windows for dividing the sound data. Also, when the number of windows is plural, consecutive windows may overlap each other.

For example, referring to FIG. 4 , sound data may be divided into a plurality of windows having 0 seconds as a starting point, 120 seconds as an ending point, and a size of 2.5 seconds. Specifically, looking at the time interval of 37.5 seconds to 40.05 seconds, which is a part of the sound data, the kth window is 37.50 seconds to 40.00 seconds, the k+1th window is 37.55 seconds to 40.05 seconds, the k+2 window is 37.60 to 40.10 seconds, The k+3th window may correspond to 37.65 seconds to 40.15 seconds, and the k+4th window may correspond to 37.70 seconds to 40.20 seconds. As such, the successive windows may overlap each other, and the overlapping degree may be determined according to the resolution of the sound data, the sliding degree of the successive windows, the window size, and the number of windows. For example, in FIG. 4 , the resolution of the acoustic data may be 0.05 seconds, the sliding degree may be 0.05 seconds, the window size may be 2.5 seconds, and the overlapping degree of successive windows may be 2.45 seconds. On the other hand, the resolution, sliding degree, and window size of the sound data are not limited to the above-mentioned numerical values, the resolution is about 0.01 seconds to about 2.00 seconds, the sliding degree is about 0.01 seconds to about 5.00 seconds, and the window size is about 0.05 It can be determined between seconds and about 5.00 seconds. More preferably, the resolution may be determined between about 0.05 seconds and about 1.00 seconds, the sliding degree between about 0.05 seconds and about 2.50 seconds, and the window size between about 0.10 seconds and about 3.00 seconds.

In consideration of the data merging process to be described later, as the overlapping degree of successive windows increases, the accuracy of data may increase, but the data processing speed may decrease due to an increase in the amount of data processing. Therefore, it is necessary to determine the overlapping degree of successive windows in consideration of the priority between data accuracy and data processing speed.

As another example, referring to FIG. 5 , the sound data is divided into a plurality of windows having 0 seconds as a starting point, 120 seconds as an ending point, and having a size of 2.5 seconds. Unlike FIG. 4 , successive windows do not overlap each other. it may not be Specifically, looking at a time interval of 37.5 seconds to 45 seconds, which is a part of the sound data, the k-th window is from 37.50 seconds to 40.00 seconds, the k+1th window is from 40.00 seconds to 42.50 seconds, and the k+2 window is from 42.50 seconds to 45.00 seconds. can correspond to When sound data is divided into windows so that successive windows do not overlap, accuracy of data may be lowered in a data merging process to be described later, but data processing speed may be reduced and data processing speed may be relatively fast.

When sound data is divided into a plurality of windows, whether successive windows overlap and, if overlapped, the degree of overlap may vary depending on target data to be obtained from the sound data.

As an example, in predicting the urine velocity in the urination process using the acoustic data, the predicted urine velocity value divides the acoustic data into a plurality of windows because high accuracy is required. It may be desirable to improve

As another example, in classifying a urination process into a urination section and a non-urination section using the sound data, it is relatively easy to determine whether urination/non-urination is present and the accuracy is generally high. However, in the process of predicting the above-mentioned yaw speed, it is desirable to increase the window sliding degree to be lower than the overlapping degree of successive windows, or to improve the data processing speed by not overlapping successive windows as shown in FIG. can

Hereinafter, for convenience of explanation, in the process of predicting the urine speed, sound data is divided into a plurality of windows so that consecutive windows overlap, and in the process of classifying the urination section and the non-urination section, it is more continuous than in the process of predicting the urine speed. A case in which the acoustic data is divided into a plurality of windows so that the overlapping degree of the windows to be overlapped is low or the continuous windows are not overlapped is mainly described, but the technical spirit of the present specification is not limited thereto, and the process of predicting the yaw velocity The sound data may be divided into a plurality of windows so that the overlapping degree of successive windows is the same for both the process of classifying the urination section and the non-urination section, and conversely, in both processes, the sound data is divided into a plurality of windows so that the continuous windows do not overlap. Of course, it can also be separated by a window.

특징 값 추출Feature value extraction

6 is a diagram illustrating a process of extracting a feature value from sound data according to an embodiment of the present specification. Here, the feature value is mainly described as being extracted from the image data of the spectrogram, but it goes without saying that extracting other types of feature values described above may also be included in the technical spirit of the present specification.

Acoustic data may generally include amplitude values over time. The sound data may be converted into spectral data including magnitude values according to frequency through processing. Here, the spectral data uses a Fourier Transform (FT), a Fast Fourier Transform (FFT), a Discrete Fourier Transform (DFT), and a Short Time Fourier Transform (STFT). can be obtained by

The spectrogram image may be obtained using the above-described acoustic data and spectrum data. Here, the spectrogram image may be a Mel-spectrogram image to which Mel-scale is applied.

The feature extractor 1200 may extract data from the spectrogram image. Data extraction may be understood as a process in which spectrogram image values corresponding to each of a plurality of windows dividing sound data are extracted.

분석을 위한 대상 데이터 획득Acquire target data for analysis

The target data may refer to a set of data generated by extracting a feature value from the converted sound data. For example, a plurality of pieces of target data may be generated from a spectrogram image of the sound data, and the target data may be understood as a set of a plurality of pieces of target data.

The number of target data, that is, the number of pieces of target data included in the target data may be the same as the number of windows. For example, referring to FIG. 7 , when sound data is divided into first to m-th windows (m is a natural number greater than or equal to 2), the target data is a first to m-th engraving target corresponding to each of the first to m-th windows. It may contain data. For another example, when the sound data is divided into first to n-th windows (n is a natural number greater than or equal to 2), the target data may include first to n-th piece target data corresponding to each of the first to n-th windows. have.

On the other hand, the number of pieces of target data included in the target data may not be the same as the number of windows. For example, the target data may include a number of pieces of target data greater than the number of windows.

The size of each piece of data to be sliced may be determined according to the unit time (or resolution) of the time axis of the spectrogram, the number of unit sections on the frequency axis, and the size of the window. For example, assuming that the data size for one slice is z*p, if the time axis unit time of the spectrogram is 0.05 seconds, the number of unit sections on the frequency axis is 512, and the window size is 2.5 seconds, z=2.5/0.05= 50, p=512.

In a method of dividing sound data into a plurality of windows in the process of generating target data, a method of dividing consecutive windows so as to overlap as shown in FIG. A classification method may be used.

The number of target data used in a urine velocity prediction model or a urination/non-urination classification model to be described later may be different from each other. For example, the number of target data used in the urine velocity prediction model may be greater than the number of target data used in the urination/non-urination classification model.

In this case, the target data used for the urine velocity prediction model and the target data used for the urination/non-urination classification model may be respectively generated from the spectrogram image. Here, the size of each piece of target data included in the target data used in the urinary velocity prediction model and the size of each piece of target data included in the target data used in the urination / non-urination classification model are the same, but the sizes used in each model are the same. The total number of pieces of target data may be different.

Alternatively, at least a portion of the target data generated from the spectrogram image may be used for the urine velocity prediction model, and at least a portion may be used for the urination/non-urination classification model. For example, the entire first to mth slice target data generated from the spectrogram image is used for the yaw velocity prediction model, and the first to nth slice target data among the first to mth slice target data (n is 2 or more and , less than m) can be used for the voiding/non-voiding classification model. Here, the first to m-th engraving target data correspond to each of the first to m-th windows in which the acoustic data or spectrogram image data is divided so that successive windows overlap each other, and the first to n-th engraving target data are the first to m-th windows. The overlapping degree of the mth window may be smaller than the overlapping degree of successive windows in the yaw velocity prediction model, or may correspond to each of the n pieces of data that do not overlap each other.

As described above, by varying the number of pieces of target data used in the urine velocity prediction model and the urination/non-urination classification model, the accuracy in the urine velocity prediction model is increased, and the data processing speed in the urination/non-urination classification model can be improved. have.

On the other hand, the number of pieces of target data used in the urine velocity prediction model or the urination/non-urination classification model may be the same. For example, the first to mth piece target data generated from the spectrogram image may be input to each of the urine velocity prediction model and the urination/non-urination classification model.

In addition, according to the learning method, the shape of the slice target data used in the urine velocity prediction model (ex. data size, total number, etc.) and the shape of the slice target data used in the urination/non-urination classification model may be different. .

요속 예측 방법How to predict yaw velocity

Hereinafter, a process of obtaining candidate yaw velocity data from target data will be described with reference to FIGS. 9 and 10 .

Referring to FIG. 9 , the yaw velocity prediction model may obtain first to mth fragment yaw velocity data by receiving first to mth slice target data.

The yaw velocity prediction model is a model trained using machine learning, and its algorithms include k-Nearest Neighbors, Linear Regression, Logistic Regression, and Support Vector Machine (SVM: Support). At least one of a vector machine, a decision tree, a random forest, or a neural network may be used. Here, as a neural network, one of Artificial Neural Network (ANN), Time Delay Neural Network (TDNN), Deep Neural Network (DNN), Convolution Neural Network (CNN), Recurrent Neural Network (RNN), or Long short-term Memory (LSTM). At least one may be selected.

For example, referring to FIG. 9 , the yaw velocity prediction model may include an input layer, a hidden layer, and an output layer.

Here, the input layer may sequentially receive the first to mth fragment target data with an input size of z*p. Alternatively, the input layer may have an input size of m*z*p and receive the first to mth piece target data at once.

Here, when the yaw velocity prediction model uses a CNN algorithm, the hidden layer may include a convolution layer, a pooling layer, and a fully-connected layer.

Here, the output layer may output fragment yaw velocity data including yaw velocity values. In this case, the fragment yaw velocity data output from the output layer may correspond to a window corresponding to the input target data. For example, when the first fragment target data corresponding to the first window is input to the yaw velocity prediction model and the first fragment yaw velocity data is output, the first fragment yaw velocity data may include yaw velocity values corresponding to the first window. have. Specifically, the first piece of yaw velocity data may include predicted yaw velocity values per unit time interval in the first window, that is, from the start point of the sound data for a time equal to the window size, and the number is the size or It may be determined according to the time axis unit time of the spectrogram (ex. If the time axis unit time of the spectrogram is 0.05 seconds and the window size is 2.5 seconds, the number of predicted yaw velocity values included in one piece of yaw velocity data is 2.5/0.05 = 50 pieces).

On the other hand, the urine velocity prediction model performs a function of outputting predicted urine velocity values, and unlike a urination/non-urination classification model to be described later, a step function, a sigmoid function, or a softmax function function) or an activation function or classifier.

The urination information extraction unit 1500 may generate candidate urine velocity data by processing the piece of urine velocity data. Here, the candidate urine velocity data may include predicted urine velocity values according to time in the recorded urination process. Each predicted yaw velocity value included in the candidate yaw velocity data may be obtained from a plurality of pieces of yaw velocity data. Specifically, the predicted yaw velocity value of a specific time interval in the candidate yaw velocity data may be obtained by using predicted yaw velocity values corresponding to the specific time interval in each of the entire piece of yaw velocity data (eg, using an average value or an intermediate value).

As an example, referring to FIGS. 4 and 10 , a part of the data output through the yaw velocity prediction model is referred to as kth to k+3 piece yaw velocity data, and the k th fragment yaw velocity data is in the k th window of 37.50 seconds to 40.00 seconds. Correspondingly, the k+1th fragment yaw velocity data corresponds to the k+1th window of 37.55 seconds to 40.05 seconds, the k+2th fragment yaw velocity data corresponds to the k+2th window from 37.60 seconds to 40.10 seconds, and k+ When the three-piece yaw velocity data corresponds to the k+3th window of 37.65 seconds to 40.15 seconds, the predicted yaw velocity values corresponding to 37.65 seconds to 37.70 seconds among the candidate yaw velocity data for the acoustic data are from 37.65 seconds to 37.70 seconds in the k-th piece yaw velocity data. The predicted yaw velocity value corresponding to the second (k,4), the predicted yaw velocity value corresponding to 37.65 seconds to 37.70 seconds in the k+1 piece yaw velocity data (k+1,3), and 37.65 seconds in the k+2 piece yaw velocity data A predicted yaw velocity value corresponding to 37.70 seconds in (k+2,2), a predicted yaw velocity value corresponding to 37.65 seconds to 37.70 seconds in the k+3 piece yaw velocity data (k+3,1), and the k th slice yaw velocity data It may be calculated as an average value of predicted yaw velocity values corresponding to 37.65 seconds to 37.70 seconds in each of the previous pieces of yaw velocity data.

In the above description, the case in which the target data used in the yaw velocity prediction model is obtained based on acoustic data divided into a plurality of overlapping windows has been described, but the target data used in the yaw velocity prediction model includes a plurality of windows that do not overlap each other It may be obtained based on the sound data separated by . In this case, the candidate yaw velocity data may be obtained by concatenating a data concatenation operation, which will be described later, for pieces of yaw velocity data that do not overlap each other in a time band.

Meanwhile, the candidate yaw velocity data may be generated by a method different from the above-described method. For example, the candidate yaw velocity data may be generated using spectral data obtained by converting the acoustic data. Specifically, the acoustic data is divided into a plurality of non-overlapping time windows to obtain fragment spectrum data corresponding to each time window, a frequency band of the fragment spectrum data is divided into a plurality of frequency windows, and RMS in each frequency window Extracting a value as a feature value, obtaining a yaw velocity prediction value in a corresponding time window using the feature values in each frequency window, and obtaining the yaw velocity prediction values obtained in each of a plurality of time windows as candidate yaw velocity data .

배뇨/비-배뇨 분류 방법How to classify urination/non-urination

Referring to FIG. 11 , the urination/non-urination classification model may obtain first to nth piece classification data by receiving first to nth piece target data.

A urination/non-urination classification model is a model trained using machine learning, and its algorithm uses at least one of k-nearest neighbors, linear regression, logistic regression, support vector machine, decision tree and random forest, or neural network. Available. Here, at least one of ANN, TDNN, DNN, CNN, RNN, and LSTM may be selected as the neural network. The voiding/non-urination classification model may have the same structure as the aforementioned urine velocity prediction model.

For example, referring to FIG. 11 , the urination/non-urination classification model may include an input layer, a hidden layer, and an output layer.

Here, the input layer may sequentially receive first to n-th piece target data with an input size of z*p. Alternatively, the input layer may have an input size of n*z*p and receive the first to nth piece target data at once.

Here, when the urination/non-urination classification model uses a CNN algorithm, the hidden layer may include a convolutional layer, a pooling layer, and a fully connected layer.

Here, the output layer may output fragment classification data including classification values. In this case, the fragment classification data output from the output layer may correspond to a window corresponding to the input fragment target data. For example, when the first piece of target data corresponding to the first window is input to the urination/non-urination classification model and the first piece of classification data is output, the first piece of classification data is predictive classification corresponding to the first window It can contain values. Specifically, the first fragment classification data may include prediction classification values per unit time interval for a time corresponding to the window size from the start point of the first window, that is, the sound data, and the number is the size of the fragment target data described above. Alternatively, it may be determined according to the time axis unit time of the spectrogram (ex. If the time axis unit time of the spectrogram is 0.05 seconds and the window size is 2.5 seconds, the number of prediction classification values included in one piece of classification data is 2.5/ 0.05 = 50 pieces).

Predicted classification values output through the urination/non-urination classification model may be values indicating a probability of a urination section or a non-urination section. For example, it may be understood that the predicted classification values have a value between 0 and 1, and the closer to 1, the higher the probability of the urination section, and the closer to 0, the higher the probability of the non-urination section. At this time, the urination information extraction unit 1500 described later uses an activation function or a classifier such as a step function, a sigmoid function, or a softmax function to generate predicted classification values or a value generated by using the predicted classification values in the urination section. It can be changed to a value indicating a value or a value indicating a non-urination interval.

Meanwhile, the predicted classification values output through the urination/non-urination classification model may be either a value indicating a urination interval (ex. 1) or a value indicating a non-urination interval (ex. 0). In order to provide the predicted classification value in the form of a value indicating a specific class as described above, the urination/non-urination classification model may use a softmax function or a softmax classifier, unlike the aforementioned urine velocity prediction model.

The urination information extraction unit 1500 may generate urination classification data by processing a plurality of pieces of classification data.

First, the urination information extraction unit 1500 may process a plurality of pieces of classification data to generate urination determination data. The urination presence/absence determination data may include values regarding whether urination/non-urination according to time in the recorded urination process. Here, the urination/non-urination-related values included in the urination presence/absence determination data may be understood as a probability value of a urination section or a probability value of a non-urination section. Each of the values included in the urination presence determination data may be obtained from a plurality of pieces of classification data. Specifically, a value corresponding to a specific window (or a specific time interval) in the urination presence or absence determination data uses prediction classification values corresponding to a specific window (or a specific time interval) in each of the entire fragment classification data (eg, an average value or a median value). value) can be obtained.

As an example, referring to FIG. 12 , when the window size for separating the acoustic data is 2.5 seconds and the sliding degree is 1.25 seconds, the kth to k+3 pieces that are a part of the data output through the urination/non-urination classification model The classification data may correspond to 37.50 seconds to 40.00 seconds, 38.75 seconds to 41.25 seconds, 40.00 seconds to 42.50 seconds, and 41.25 seconds to 43.75 seconds, respectively, and the value in a specific window of the urination determination data is specific in each of the slice classification data. It may be obtained as an average value or a median value of prediction classification values corresponding to the window.

The urination presence determination data may be used to determine the presence or absence of a urination section, which will be described later.

The urination information extraction unit 1500 may obtain urination classification data for the sound data by processing the urination presence/absence determination data. For example, the urination information extractor 1500 applies a step function, a sigmoid function, or a softmax function to the urination presence or absence determination data to indicate a urination section value (ex. 1) Alternatively, urination classification data including values (ex. 0) indicating a non-urination interval may be generated.

The urination information extraction unit 1500 may generate urination classification data by processing the fragment classification data. Here, the urination classification data may include predicted classification values according to time in the recorded urination process. Each of the prediction classification values included in the urination classification data may be obtained by concatenating a plurality of pieces of classification data. Specifically, the prediction classification value of a specific time interval in the urination classification data may be obtained as a prediction classification value corresponding to the specific time interval in the fragment classification data corresponding to the window including the specific time interval. Here, the predicted classification value may be a probability value related to a urination section or a non-urination section (eg, a value between 0 and 1, which may be understood to be a higher probability of a urination section as the value is closer to 1). Alternatively, the predicted classification value may be a value indicating a urination interval (ex. 1) or a value indicating a non-urination interval (ex. 0).

As an example, referring to FIGS. 5 and 13 , a part of the data output through the urination/non-urination classification model is referred to as kth to k+2 fragment classification data, and the kth fragment classification data is from 37.50 seconds to 40.00 seconds. Corresponds to the kth window, the k+1th fragment classification data corresponds to the k+1th window of 40.00 seconds to 42.50 seconds, and the k+2th fragment classification data corresponds to the k+2th window from 42.50 seconds to 45.00 seconds. At this time, the prediction classification values corresponding to 37.50 seconds to 40.00 seconds among the urination classification data for the acoustic data are the prediction classification values (0,...,1,...,1) of the k-th piece classification data, 40.00 seconds The prediction classification values corresponding to 42.50 seconds in , are the prediction classification values (1,...,1,1...1) of the k+1th piece classification data, and the prediction classification values corresponding to the 42.50 seconds to 45.00 seconds are It may be obtained as predicted classification values (1,0,0,1,...,0) of the k+2th piece of classification data.

Meanwhile, the urination classification data may be generated by a method other than the above-described method.

For example, the urine classification data may be generated using spectrum data obtained by converting sound data. Specifically, the acoustic data is divided into a plurality of time windows to obtain fragment spectrum data corresponding to each time window, a frequency band of the fragment spectrum data is divided into a plurality of frequency windows, and the RMS value in each frequency window is characterized It is extracted as a value, and it is determined whether the corresponding time window is a urination section or a non-urination section using the feature values in each frequency window, and the result can be obtained as urination classification data.

As another example, the urination classification data may be generated using sound data. Specifically, the zero-crossing rate is extracted as a feature value for the acoustic data, a urination section and a non-urination section are determined in the acoustic data based on the extracted feature value, and the result can be obtained as urination classification data. .

배뇨 데이터 획득 방법How to acquire urination data

Hereinafter, a method of generating urination data will be described with reference to FIGS. 14 and 15 .

The urination information extractor 1500 may calculate urination data by using the candidate urine velocity data and the urination classification data. The urination data may be obtained in a form including time-dependent predicted values of urination for a recorded urination process. For example, referring to FIG. 14 , urination data may be obtained by convolving candidate urine velocity data and voiding classification data. For another example, the urination data may be obtained by multiplying each of the time-dependent urine velocity prediction values included in the candidate urine velocity data by each of the temporal classification prediction values included in the urination classification data. In this case, the urination data may be expressed as a matrix of discrete values.

The urination information extraction unit 1500 may extract urination information by using the obtained urination data. For example, the urination information extraction unit 1500 obtains the largest value among the predicted urine velocity values included in the obtained urination data as the maximum urine velocity value, and calculates the voiding amount or total voiding amount for each section through discrete integration. can do. On the other hand, referring to FIG. 14 , the urination information extraction unit 1500 may obtain a urination graph for the recorded urination process by processing the urination data (eg, smoothing, interpolation, etc.), and the maximum urine velocity value from the urination graph, It is possible to calculate the average urine velocity value, urination time, voiding amount for each section and total voiding amount.

The urination data may be obtained by processing the candidate urine velocity data and the urination classification data, respectively, and then performing an operation on the processed data. For example, referring to FIG. 15 , the urination information extraction unit 1500 processes the candidate urine velocity data to generate a candidate urine velocity graph for the recorded urination process, and processes the urination classification data to urinate / for the recorded urination process. A non-urination section graph is generated, and a urination graph for the recorded urination process can be obtained by performing a convolution operation on the generated candidate urine velocity graph and the urination/non-urination section graph. Here, in the process of processing the candidate urine velocity data or in the process of processing the urination classification data, the operation of smoothing or interpolation may be performed.

음향 분석 방법Acoustic analysis method

Hereinafter, an acoustic analysis method for obtaining urination information for a urination process using the acoustic analysis system 1000 will be described in stages with reference to FIGS. 16 and 17 , but overlapping with the above-described parts will be omitted. .

배뇨 구간 유무 판단 없이 음향 분석 진행Acoustic analysis proceeds without judging the presence or absence of a urination section

Referring to FIG. 16 , the acoustic analysis method includes: acquiring acoustic data (S110), processing the acquired acoustic data (S120), estimating the urinary velocity (S130), and classifying a urination/non-urination section It may include a step (S140), a step of obtaining urination data (S150), and a step of outputting urination information (S160).

The acoustic analysis system 1000 may acquire acoustic data (S110). The acoustic analysis system 1000 may acquire acoustic data recorded by a human urination process from the recording apparatus 2000 . Alternatively, the acoustic analysis system 1000 may acquire arbitrary acoustic data from the outside, but the corresponding acoustic data may not be a recording of a urination process.

The acoustic analysis system 1000 may process the acquired acoustic data (S120). The acoustic analysis system 1000 may process the acquired acoustic data using the preprocessor 1100 and the feature extraction unit 1200 . As for the method of processing the sound data, the detailed description thereof will be omitted as described above.

The acoustic analysis system 1000 may predict the yaw velocity (S130). The acoustic analysis system 1000 obtains a piece urine velocity data group corresponding to each window in which the acoustic data is separated from the processed sound data using the urine velocity prediction unit 1300 , and uses the urination information extraction unit 1500 . to obtain candidate yaw velocity data from the fragment yaw velocity data group. The candidate yaw velocity data may include values predicted by yaw velocity in the overall acoustic data. Since the process of generating the candidate yaw velocity data has been described above, specific details thereof will be omitted.

The acoustic analysis system 1000 may classify a urination section and a non-urination section ( S140 ). The acoustic analysis system 1000 obtains a fragment classification data group corresponding to each window in which the acoustic data is separated from the acoustic data processed data using the urination/non-urination classification unit 1400, and a urination information extraction unit ( 1500) to obtain urination classification data from the fragment classification data group. The urination classification data may include classification values for classifying a urination section and a non-urination section in the overall sound data. As for the process of generating the urination classification data, the above-described details will be omitted.

Predicting the above-described urine speed (S130) and classifying the urination/non-urination section (S140) may be performed in parallel or sequentially.

The acoustic analysis system 1000 may acquire urination data (S150). The acoustic analysis system 1000 may acquire urination data from the candidate urine velocity data and the urination classification data by using the urination information extractor 1500 . The urination data may be understood as including the urine velocity prediction values selected by the urination classification data from among the urine velocity prediction values included in the candidate urine velocity data in the acoustic data. As for the process of generating urination data, the above-described details will be omitted.

The acoustic analysis system 1000 may output urination information (S160). The acoustic analysis system 1000 extracts urination information from urination data using the urination information extraction unit 1500 , and provides the extracted urination information to the user through the output unit 1700 or the recording device 2000 and/or It can be provided to the external server (3000). Here, the output urination information may include a voiding amount, a maximum urine velocity, an average urine velocity, and a urination time as shown in FIG. 3 .

배뇨 구간 유무 판단 후 음향 분석 진행Acoustic analysis proceeds after determining whether there is a urination section

On the other hand, the urination information analysis method may perform a judgment on the presence or absence of urination prior to the prediction of the urine velocity in order to improve the efficiency in the data processing process.

Referring to FIG. 17 , the method for analyzing urination information includes: acquiring acoustic data (S210), processing the acquired acoustic data (S220), classifying urination/non-urination sections (S230), presence or absence of a urination section It may include a step of determining (S240), a step of predicting a urine speed (S250), a step of obtaining urination data (S260), and a step of outputting urination information (S270). Here, the step of obtaining the sound data (S210), the step of processing the obtained sound data (S220), and the step of outputting the urination information (S270) have been described above in FIG. 16 , so the duplicated content will be omitted.

The acoustic analysis system 1000 may classify the urination/non-urination section prior to predicting the urinary velocity ( S230 ). The acoustic analysis system 1000 may acquire urination classification data from the data processed by the acoustic data using the urination/non-urination classification unit 1400 and the urination information extraction unit 1500 .

The acoustic analysis system 1000 may determine whether there is a urination section (S240). The acoustic analysis system 1000 may determine whether the acoustic data is data recorded with a urination process using the urination classification data obtained in the urination/non-urination classification step S230 . For example, the sound analysis system 1000 may determine whether a urination section exists between a start point and an end point of the sound data.

As an example, the acoustic analysis system 1000 may determine whether a urination process is reflected in the acoustic data based on the above-described urination presence or absence determination data. For convenience of explanation, a value included in the urination presence/absence determination data is referred to as a determination value. The acoustic analysis system 1000 may determine that the urination process is reflected in the acoustic data when at least one of the determination values included in the urination presence determination data is greater than or equal to a threshold value. Alternatively, the acoustic analysis system 1000 determines that the urination process is reflected in the acoustic data when the ratio of the judgment value equal to or greater than the threshold value among the judgment values included in the urination determination data is a certain ratio (ex. 5% to 50%) or more. can judge A threshold value for determining whether there is a urination section may be set in various ways. For example, the threshold value may be determined between 0.30 and 0.95. More preferably, the threshold value may be determined between 0.50 and 0.80.

As another example, the acoustic analysis system 1000 may determine whether a urination process is reflected in the acoustic data based on the urination classification data. Specifically, when the number of classification values indicating the urination section in the urination classification data is greater than or equal to a predetermined number, the acoustic analysis system 1000 may determine that the sound related to the urination process is reflected in the acoustic data. Alternatively, the acoustic analysis system 1000 calculates a ratio of the number of classification values indicating a urination section to the number of classification values indicating a non-urination section in the urination classification data, and if the calculated ratio is greater than a certain ratio, the urination section It can be judged that there is

The acoustic analysis system 1000 may generate only urination determination data or both urination presence determination data and urination classification data according to the above-described determination method in step S240 of determining the presence or absence of a urination section.

When it is determined that there is a urination section in the acoustic data, the acoustic analysis system 1000 may enter the urination velocity prediction step S260 , and when it is determined that there is no urination section in the acoustic data, the acoustic analysis may be terminated. When it is determined that there is no urination section in the acoustic data, the acoustic analysis system 1000 may not generate urination classification data from the urination presence/absence determination data when determining whether there is a urination section using the urination presence/absence determination data.

When it is determined that a urination section exists in the acoustic data, the acoustic analysis system 1000 may predict the urine velocity (S250). The acoustic analysis system 1000 obtains a piece urine velocity data group corresponding to each window in which the acoustic data is separated from the processed sound data using the urine velocity prediction unit 1300 , and uses the urination information extraction unit 1500 . to obtain candidate yaw velocity data from the fragment yaw velocity data group. The candidate yaw velocity data may include values predicted from the overall acoustic data, and the process of generating the candidate yaw velocity data will be omitted as described above.

Here, the processed acoustic data used in the urine velocity prediction unit 1300 may be different from the data used in the urination/non-urination section classification step S230 . For example, the data used by the urine velocity prediction unit 1300 is data corresponding to each window when the sound data is divided into m windows so that consecutive windows are overlapped, and in the urination/non-urination section classification step (S230) Data used is data corresponding to each window when sound data is divided into n windows so that consecutive windows do not overlap, and n may be smaller than m.

The acoustic analysis system 1000 may acquire urination data (S260). The acoustic analysis system 1000 may acquire urination data by using the urination classification data obtained in the urination/non-urination section classification step S230 and the candidate urine velocity data obtained in the urine velocity prediction step S250 .

As described above, in the urination information analysis method, if the urination-related sound is not reflected in the sound data to be analyzed by determining the presence or absence of a urination section prior to performing the urine velocity prediction step (S250), it is necessary to analyze the data in advance. can be prevented in

음향 분석 시스템 효과Acoustic analysis system effect

Hereinafter, in the method for analyzing urination information using the acoustic analysis system 1000 , a practical effect when the urination/non-urination classification process is performed will be described.

When the urination/non-urination classification model is not used in the urination information analysis method, the amount of urination may be excessively predicted compared to the case where the urination/non-urination classification model is used. For example, in (a) to (d) of FIG. 18 , the urination/non-urination classification model was measured to be larger than the case in which the urination/non-urination classification model was not applied. In particular, in the case of (a) and (c) of FIG. 18, it can be seen that the urination amount differs by about two times between the case where the urination/non-urination classification model is applied and the case where the urination classification model is not applied.

In the case of not using the urination/non-urination classification model in the urination information analysis method, the maximum urinary velocity may be overestimated than when the urination/non-urination classification model is used. For example, in (a) and (c) of FIG. 18 , the maximum urinary velocity (Qmax) was measured to be greater when the voiding/non-urination classification model was not applied than when the urination/non-urination classification model was not applied. For men, the maximum urinary velocity value is a very important factor in diagnosing urination function, so it is very important not to measure the maximum urine velocity excessively.

In the case of not using the urination/non-urination classification model in the method of analyzing the voiding information, it is difficult to accurately measure the flowtime, voiding time, than when the urination/non-urination classification model is used. For example, when the urination/non-urination classification model is not applied in (a) and (c) of FIG. 18 , there are a number of peak values, making it difficult to measure the start and end points of urination and the resulting urination time. , it can be seen that when the urination/non-urination classification model is applied, the start point and the end point are clear, so that the total urination time can be accurately calculated.

As described above, in order to accurately measure voiding information such as the amount of voiding, the maximum urinary velocity, and the voiding time, it is absolutely essential to use a voiding/non-urination classification model.

모델 학습 방법How to train a model

Hereinafter, a process of learning the urinary velocity prediction model and the urination/non-urination classification model described above with reference to FIGS. 19 and 20 will be described in detail.

요속 예측 모델 학습 방법How to train a yaw velocity prediction model

Referring to FIG. 19 , the urine velocity prediction model learning method includes the steps of collecting acoustic data in the urination process ( S310 ), processing the collected acoustic data to generate characteristic data ( S320 ), and a urine velocity value at a preset cycle in the urination process. Measuring (S330), generating measured yaw velocity data using the measured yaw velocity value (S340), generating a learning data set by labeling the measured yaw velocity data in the feature data (S350), and a learning data set It may include generating a yaw velocity prediction model using (S360).

Each step is described below. Each step of the yaw velocity prediction model learning method may be performed by a human or a separate processor.

In the course of learning the urinary velocity prediction model, acoustic data for the urination process may be collected (S310). The acoustic data may be obtained by recording a human urination process. Meanwhile, in the acoustic data collection step ( S310 ), acoustic data that does not include a human urination process may be collected in order to improve the performance of the urine velocity prediction model. The above-described recording apparatus 2000 may be used to collect sound data.

Feature data may be generated from the acoustic data collected in the course of learning the yaw velocity prediction model (S320). The feature data may refer to data including feature values extracted from the sound data. Here, the feature value is a spectrum time domain spectrum magnitude value, a frequency band spectrum magnitude value, a frequency band root mean square value, a spectrogram magnitude value, a Mel-spectrogram magnitude value, Bispectrum Score, Non-Gaussianity Score, Formants Frequencies, Log Energy , Zero Crossing Rate, Kurtosis, and Mel-frequency cepstral coefficient may be included.

The feature data may be divided into a plurality of windows. For example, the feature data may include a plurality of pieces of feature data corresponding to each of a plurality of windows continuously determined between a start point and an end point of the sound data or feature data.

Since the process of generating the feature data and the engraving feature data is the same as the process of generating the above-described target data and the engraving target data, a detailed description thereof will be omitted.

In the course of learning the urine velocity prediction model, the urine velocity value may be measured at a preset period for the urination process ( S330 ). For example, a urine velocity value in the process of urination may be measured using a scale. For example, during urination, the urine velocity value may be measured using a toilet with a built-in scale or a toilet mounted on the scale. Specifically, as the urination process is performed through a toilet with a built-in scale or a toilet mounted on the scale, the amount of urination over time can be measured using the change over time of the weight measured by the scale, and from the data on the amount of urination over time A yaw velocity value over time can be measured. In this case, the period in which the urine velocity value is measured or the period in which the amount of urination is measured may correspond to the resolution of the sound data.

The measured yaw velocity data may be generated using the yaw velocity value measured in the yaw velocity measurement step ( S330 ) ( S340 ). The measured yaw velocity data may include yaw velocity values measured or calculated at a preset period.

The measured yaw velocity data may be divided into a plurality of windows. For example, the measured urine velocity data may include a plurality of pieces of measured urine velocity data corresponding to each of a plurality of windows continuously determined between a measurement start time and a measurement end time of the urination process. In this case, the size of the window dividing the measured yaw velocity data may be the same as the size of the window dividing the above-described characteristic data. In addition, the number of pieces of measured yaw velocity data included in the measured yaw velocity data may be the same as the number of pieces of measured yaw velocity data included in the aforementioned feature data.

Steps (S310, S320) of collecting the above-described acoustic data and generating characteristic data from the collected acoustic data are steps of measuring a urine velocity value in the urination process and generating actual measured urine velocity data using the measured urine velocity value (S330, S340) ), it may be performed at a previous time point or a later time point based on it, or it may be performed in parallel.

In the course of learning the yaw velocity prediction model, the measured yaw velocity data may be labeled with the feature data to generate a training data set (S350). The feature data and the measured yaw velocity data may be labeled with data corresponding to each window section. For example, the first piece characteristic data included in the feature data and corresponding to the first window section is included in the measured yaw velocity data, and is labeled with the first piece measured yaw velocity data corresponding to the first window section to form the first learning data can be created.

The yaw velocity prediction model may be generated using the aforementioned training data set (S360). The yaw velocity prediction model may be trained using a training data set. The learned yaw velocity prediction model may be used to obtain the fragment yaw velocity data using the above-described fragment target data.

배뇨/비-배뇨 분류 모델 학습 방법How to train a urination/non-urination classification model

Referring to FIG. 20 , the urination/non-urination classification model learning method includes the steps of collecting acoustic data in the urination process (S410), processing the collected acoustic data to generate characteristic data (S420), and in the urination process in advance. Determining whether urination or non-urination in a set cycle (S430), generating urination/non-urination determination data (S440), labeling the urination/non-urination determination data in the feature data to generate a learning data set It may include step S450 and generating a urination/non-urination classification model using the training data set (S460).

Each step is described below. Each step of the urination/non-urination classification model training method may be performed by a human or a processor. Here, the step (S410) of collecting the sound data and the step (S420) of generating the characteristic data by processing the collected sound data are the same as the steps (S310) and (S320) described in FIG. 19 , and specific details are omitted. let it do

In the urination/non-urination classification model learning process, it may be determined whether urination or non-urination with respect to the urination process ( S430 ). For example, a urination or non-urination section may be determined with respect to the urination process using sound data recorded during the urination process. Specifically, the intensity in each time section in the sound data recorded with the urination process may be compared with a preset value to determine the urination section or the non-urination section. For another example, a urination section and a non-urination section may be determined in the urination process by using the voiding amount data obtained by measuring the urination amount over time in the urination process. In measuring the amount of urination for the urination process, the above-described method for measuring the amount of urination may be used.

Urination/non-urination determination data may be generated using the urination section and the non-urination section determined in the urination or non-urination determination step ( S430 ) ( S440 ). The urination/non-urination determination data may include determination values determined at a preset period in the time domain.

The urination/non-urination determination data may be divided into a plurality of windows. For example, the urination/non-urination determination data may include a plurality of pieces of determination data corresponding to each of a plurality of windows continuously determined between a measurement start time and a measurement end time of the urination process. In this case, the size of the window dividing the urination/non-urination determination data may be the same as the size of the window dividing the characteristic data generated in step S420 . Also, the number of pieces of piece determination data included in the urination/non-urination determination data may be the same as the number of pieces of piece characteristic data included in the feature data generated in step S420 .

Steps (S410, S420) of collecting the above-described acoustic data and generating characteristic data from the collected acoustic data is a step of determining whether urination or non-urination in the urination process and generating judgment data using the judgment value (S430) , S440) may be performed at an earlier time or a later time, or may be performed in parallel.

In the process of training the urination/non-urination classification model, the urination/non-urination determination data may be labeled with the feature data to generate a training data set (S450). The feature data and the urination/non-urination determination data may be labeled with data corresponding to each window section. For example, the first piece of feature data included in the feature data and corresponding to the first window section is included in the urination/non-urination determination data, and is labeled with the first piece of determination data corresponding to the first window section. 1 Learning data may be generated.

The urination/non-urination classification model may be generated using the above-described training data set (S460). Specifically, the urination/non-urination classification model may be trained using a training data set. The learned urination/non-urination classification model may be used to obtain fragment classification data using the above-described fragment target data.

Hereinafter, a data correction method for correcting urination data or urination information obtained using sound data will be described. Here, the urination data or urination information corrected by applying the data correction method may be urination data or urination information obtained by other methods as well as when obtained by the method of obtaining urination information with high accuracy described through FIGS. 1 to 20. make it clear in advance that

In addition, for convenience of explanation, the following mainly describes the case where the urination data or urination information to be corrected is a predicted value regarding the amount of urination, but the technical spirit of the present specification is not limited thereto, and the above-described maximum urine velocity, average urine velocity, and error time , or a case of correcting a predicted value for urination time may be similarly applied.

배뇨 데이터 보정 방법How to correct urination data

The actual data may include measurement values of the amount of urine measured during the urination process of any one individual (eg, a general public, a patient, or a subject, etc.) for a certain period of time. For example, the actual measurement data may include a measurement value of the amount of urine measured during a person's urination process for one day. As another example, the actual measurement data may include a measurement value of a voiding amount measured in a urination process of a person for a plurality of days. The urine output measurement value may be obtained in various ways. For example, the urine output measurement value may be obtained by using a measuring device such as a measuring container such as a paper cup or a graduated cup, or a portable, household urine velocity measuring device or a urine velocity measuring device used in a medical institution, a scale, or the like.

The number of measurement values of the amount of urination included in the actual data may be the same as the number of times of urination of the subject.

The prediction data may include voiding amount prediction values calculated in the urination process of any one individual for a certain period of time. For example, the prediction data may include a urination amount prediction value obtained by recording a urination process of a person during one day and using the corresponding acoustic data. As another example, the prediction data may include a urination amount prediction value obtained by recording a urination process of a person for a plurality of days and using sound data according thereto. The urination amount predicted value may be obtained in various ways. For example, the predicted value of urination may refer to a value derived from a process of estimating a urine velocity or an amount of urination by analyzing sound data obtained by recording a human urination process by any method. Specifically, the urination amount predicted value may be obtained using the aforementioned acoustic analysis system 1000 . At this time, the urination amount predicted value is calculated from the candidate urine velocity data obtained through the urine velocity predictor 1300 and the urination information extraction unit 1500 or the urine velocity predictor 1300, the urination / non-urination classification unit 1400, and It may be calculated from urination data obtained through the urination information extraction unit 1500 .

The number of urination amount prediction values included in the prediction data may be the same as the number of urinations of the subject.

Referring to FIG. 21 , the measured data (signal Χ) and predicted data (marker ●) for a person's urination process for an arbitrary period (January 29 to February 5) are graphically shown, and the prediction The data has a relatively large value when compared with the actual data. Specifically, the measurement values for the amount of urination for the urination process from January 29 to February 1 are within about 100 mL to about 500 mL, whereas the predicted values for the amount of urine for the urination process from February 2 to February 5 are from about 300 mL to about 1000 mL. In other words, when the urination amount is predicted using the acoustic data, it may have a value greater than the actually measured urination amount, and this needs to be corrected. Hereinafter, a case in which the predicted urination amount has a larger value than that of the actual measurement is mainly described, but the technical spirit of the present specification is not limited thereto. can be applied similarly to

Referring to FIG. 22 , corrected prediction data (marked □) obtained by correcting the prediction data is displayed in the graph shown in FIG. 21 . Specifically, in FIG. 22, corrected predicted values obtained by scaling the urination amount predicted values from February 2 to February 5 are displayed, and the corrected predicted values have a value within about 100 mL to about 400 mL. can

In general, a person's urination amount is proportional to the size of the bladder, and even in different urination processes, the voiding amount value may have a value within a certain range.

In consideration of such circumstances, the predicted data for the human urination process may be corrected by using the actual data on the human urination process. For example, when the urination amount measurement values of the actual data have a first range and the urination amount prediction values of the prediction data have a second range, the urination amount prediction values of the prediction data are set such that the second range is the same as or similar to the first range. can be corrected.

On the other hand, the human urination amount may be maintained within a range of its value within a certain period. In other words, since a person's urination amount can be determined according to the size of the person's bladder, the range of a person's urination value cannot be maintained unless there are circumstances such as a change in physical characteristics, treatment, or the occurrence of a disease. can

In consideration of such circumstances, with respect to urination-related data collected at different time points, data at one time point may be corrected based on the data at one time point. For example, current or future forecast data may be corrected using past measured data, current or future forecast data may be corrected using past measured data and predicted data, and vice versa. It is possible. Meanwhile, when there is a special circumstance such as a change in the above-described physical characteristics, treatment, or occurrence of a disease, urination-related data may be collected again.

Hereinafter, with reference to FIGS. 23 to 26 , a method of correcting the predicted data for the amount of urine collected in the past or to be collected in the future using the actually measured data on the amount of urine collected will be described below.

Referring to FIG. 23 , the data correction method includes the steps of obtaining measured data ( S510 ), obtaining predicted data ( S520 ), obtaining a compensation value using the measured data and the predicted data ( S530 ), and predicting a target It may include obtaining data (S540), and correcting the target prediction data using the correction value (S550).

At least a part of each step in the data correction method may be performed through the control unit 1900 of the aforementioned acoustic analysis system 1000 or a separate processor.

For example, at least a part of the data correction method may be performed by the data correction unit. The data compensator may be provided in the form of an electronic circuit that performs a control function by processing an electrical signal in hardware, or may be provided in the form of a program or code that drives the hardware circuit in software. The data corrector may receive data from the outside and perform a data correction method using the received data or on the received data.

Hereinafter, each step will be described in detail.

In the data correction method, actual measurement data may be obtained (S510). As described above, the measured data refers to data obtained by measuring the amount of urination during the urination process of one individual. Specifically, the actual measurement data may include a measurement value of the amount of urination measured for each urination process of one person.

In operation S510 in which the measured data is acquired, a group of measured data for a plurality of entities may be acquired. For example, the above-described measured data group may include first to s-th measured data (s is a natural number equal to or greater than 2) for the first to s-th entities. Each actual measurement data may include at least one urination amount measurement value related to the urination process of the corresponding individual. As a result, acquiring the actual data group may be understood as collecting measurement values of the amount of urine measured in the urination process of a plurality of people.

In the data correction method, prediction data may be obtained (S520). As described above, the prediction data refers to data that predicts the amount of urination in the urination process of any one individual. Specifically, the prediction data may include a urination amount predicted value calculated using sound data recorded in the urination process of one person.

In operation S520 in which prediction data is obtained, prediction data groups for a plurality of entities may be obtained. For example, the prediction data group may include first to t-th prediction data (t is a natural number equal to or greater than 2, which may be the same as or different from s) for the first to t-th individuals. Each prediction data may include at least one urination amount prediction value related to the urination process of the corresponding individual. As a result, acquiring the prediction data group may be understood as collecting the urination amount prediction values calculated in the urination process of a plurality of people.

The step of obtaining the measured data ( S510 ) and the step of obtaining the prediction data ( S520 ) may be performed sequentially or in parallel.

In the data correction method, a compensation value may be obtained using measured data and predicted data ( S530 ). The compensation value may be understood as a value for correcting the collected prediction data or the prediction data to be collected later.

The compensation value may be calculated by applying a data analysis technique to the measured data acquired in the actual data acquisition step S510 and the predicted data acquired in the predicted data acquisition step S520 .

As an example, the compensation value may be calculated using a descriptive statistical analysis technique. Specifically, the compensation value is a representative value of the actual measurement data (eg, the average value, median value, mode value, variance value, and/or standard deviation value of at least some of the measurement values of urination) and a representative value of the prediction data (ex. It may be calculated by comparing an average value, a median value, a mode value, a variance value, and/or a standard deviation value for at least some of the voiding amount predicted values.

As another example, the compensation value may be calculated using a regression analysis technique. Specifically, when a value obtained from measured data is used as either an independent variable or a dependent variable, and a value obtained from prediction data is used as the other of the independent variable or dependent variable, a regression coefficient calculated through a regression analysis technique coefficient) can be obtained as a compensation value. Here, simple liner regression, multi linear regression, logistic regression, ridge regression, lasso regression, and multinomial regression are regression analysis. Polynomial regression or non-linear regression may be selected.

The reward value may be obtained using machine learning other than multi-dimensional scaling (MDS), principal component analysis (PCA), or regression analysis in addition to the above-described data analysis technique.

The compensation value may be calculated using the measured data group and the predicted data group. In other words, the compensation value may be calculated using the measured data and the predicted data for a single entity, or may be calculated using the measured data group and the predicted data group for a plurality of entities, that is, several people. A specific method of calculating the compensation value will be described later.

In the data correction method, target prediction data may be obtained (S540). The target prediction data may refer to data to be corrected using the above-described compensation value. For example, the target prediction data may refer to a urination amount predicted value calculated from acoustic data acquired before or after the data collection period in which the predicted data used to calculate the compensation value or the actual data is acquired. Of course, prediction data or urination amount prediction values used to obtain a compensation value may be target prediction data.

The target prediction data to be corrected may be prediction data for an arbitrary entity. For example, the predicted data group and the actually measured data group used to calculate the correction value include at least the first measured data and the first predicted data for the first entity, and the second measured data and the second predicted data for the second entity. When data is included, the data to be corrected may include prediction data or measurement values of voiding volume for the first and second individuals as well as for a third individual different from the first and second individuals. In other words, the above-described correction value is not limited to the prediction data of a specific entity involved in calculating the correction value, but may also be used to correct the prediction data of an entity not involved in calculating the correction value.

In the data correction method, the target prediction data may be corrected using the correction value ( S550 ). For example, a corrected urination amount predicted value may be generated by multiplying the target voiding amount predicted value and the corrected value of the target predicted data.

Hereinafter, a data correction method will be described in detail with reference to FIGS. 24 to 26 .

Referring to FIG. 24 , the urination amount prediction for the subject may be performed using the data collected during the data collection period.

In the data collection period, a group of measured data may be acquired. The measured data group is data for a plurality of entities, and may mean a set of measured data for each entity. For example, the measured data group may include first to s-th measured data for the first to s-th objects.

A group of acoustic data may be acquired in the data collection period. The sound data group is data for a plurality of entities, and may mean a set of sound data for each entity. For example, the sound data group may include first to t-th sound data for the first to t-th objects. Acoustic data for each individual may be understood to be obtained by recording a urination process during a data collection period of the subject, and may refer to data obtained by recording a single or multiple urination processes.

A group of prediction data may be obtained in the data collection period. The prediction data group is data for a plurality of entities, and may mean a set of prediction data for each entity. For example, the prediction data group may include first to t-th prediction data for the first to t-th entities.

The prediction data group may be calculated from the acoustic data group. For example, the first prediction data for the first object may be calculated from the first sound data for the first object. The first prediction data may include at least one urination amount prediction value, and the number of urination amount prediction values included in the first prediction data may correspond to the number of urination processes reflected in the first acoustic data.

The first to t-th entities related to the prediction data group may include first to s-th entities related to the actual measurement data group.

The data collection period may be a short period of time or days, or a long period of months or years. For example, the data collection period may be 6 hours, 12 hours, 18 hours, 1 day, 4 days, 7 days, 10 days, 14 days, 1 month, 3 months, 6 months, or 1 year.

Prior to calculating the correction value, a data set group may be generated from the measured data group and the predicted data group.

The data set group may include a plurality of data sets. For example, the data set group may include a data set for each of a plurality of entities.

The data set may be generated from the measured data in the measured data group and the predicted data in the predicted data group. The data set may include a value related to the measurement of the amount of urination and a value related to the prediction of the amount of urination. For example, the first data set included in the data set group is generated from the first measured data and the first predicted data, and specifically, the representative measured value and the first prediction that are representative values of the urination amount measured values included in the first measured data It may be generated by associating a representative predicted value that is a representative value of the voiding amount predicted values included in the data. In this case, the representative measured value may be a sum, average, median, mode, variance, or standard deviation value of the urination amount measured values included in the actual data, and the representative predicted value is the voiding amount predicted value included in the predicted data. It may be a sum of these values, a mean value, a median value, a mode value, a variance value, or a standard deviation value.

In other words, the data set may be understood as a concept in which at least some of the measured data and the predicted data for one individual are expressed as one set. Accordingly, one data set may be generated for one entity, or a plurality of data sets may be generated.

For example, when a data set group is generated from the actual data group and the predicted data group collected for a plurality of subjects during one day, the data set includes the average measured value for one day and the average predicted value for one day. A group can contain one data set per entity.

In another example, a group of data sets is generated from a group of ground truth data and a group of predicted data collected for a plurality of subjects over 4 days, and the data set includes the average measured value over the day and the average predicted value over the day per subject. Four data sets can be created.

The process of generating a data set is not limited to the above-described example, and a data set may be generated for each of a plurality of urination processes occurring on the same day, and one data set for all urination processes occurring during a plurality of days for each individual. Three may be created.

A data set can be expressed in various forms. For example, as shown in FIGS. 25 and 26 , the data set may be displayed in the form of coordinates on a graph.

A calibration value may be calculated from a group of data sets using a calibration process.

For example, a regression analysis technique may be used in calculating the correction value. Referring to FIG. 25 , the data set group may include a data set in which the average of the voiding amount predicted values for each individual is an independent variable and the average of the urination amount measured values for each individual is a dependent variable, and the data set group is used. Thus, a function representing the relationship between the urination amount predicted value and the urination amount measured value may be obtained. In this case, the obtained function may be a linear function, and a slope thereof may be obtained as a correction value. Also, since the predicted value of urination has a larger value than the measured value of urination, the slope may have a value between 0 and 1.

Target prediction data for the target object may be corrected using the calculated correction value. The target entity may refer to an entity for which urination information is to be acquired. The target prediction data may refer to prediction data regarding the urination process of the target object. For example, a first target predicted value may be calculated from acoustic data obtained by recording the urination process of the first target object, and the first corrected data may be obtained by applying a correction value to the first target predicted value. can Similarly, a second target predicted value may be calculated from the acoustic data obtained by recording the urination process of the second target, and the second corrected data may be obtained by applying a correction value to the second predicted target value. .

The target prediction data may be acquired after the data collection period, and the prediction data acquired during the data collection period may also be the target prediction data.

By using the correction value, the target prediction data can be corrected in a range similar to the measured value. Referring to FIGS. 25 and 26 , in FIG. 25 , in a data set of more than half of the pre-correction data set group, the voiding amount predicted value has a larger value than the voiding amount measured values and is far from the direct proportional baseline, whereas in FIG. 26 , the data after correction In the set group, most data sets can be located near the direct proportional baseline.

In the above, the process of calculating the correction value using the measured data group and the predicted data group for a plurality of entities has been described, but the technical idea of the present specification is not limited thereto, and only the measured data for one entity is used It is also possible to correct the prediction data.

As an example, actual measurement data may be collected for the target entity during the data collection period, statistical values of the collected actual measurement data may be calculated, and prediction data of the target entity may be corrected based on the calculated statistical values.

Here, the statistical value of the actually measured data may be at least one of a mode value, a median value, an average value, a variance value, or a standard deviation value of measurement values of the amount of urination measured in the urination process of the subject during the data collection period.

The predicted data for the target object may be corrected by correcting the voiding amount predicted values so that the statistical values of the voiding amount prediction values included in the prediction data for the target object correspond to the statistical values of the actually measured data for the target object.

Meanwhile, the prediction data for the target object to be corrected may be calculated from acoustic data about the urination process of the target object obtained before the data collection period, during the data collection period, or after the data collection period.

Through the above-described data correction method, the correction of the predicted data for the urination process may be performed using the measured data for the urination process or the measured data and the predicted data for the urination process. In this correction process, when the actual data collection is easy, for example, when measuring the amount of urination in the urination process using a paper cup, the corrected prediction data can secure a certain level of accuracy while the correction value is easily calculated. It can be said that it is very useful in that it can be used.

Features, structures, effects, etc. described in the above embodiments are included in at least one embodiment of the present invention, and are not necessarily limited to only one embodiment. Furthermore, the features, structures, effects, etc. illustrated in each embodiment can be combined or modified for other embodiments by those of ordinary skill in the art to which the embodiments belong. Accordingly, the contents related to such combinations and modifications should be interpreted as being included in the scope of the present specification.

In addition, although the embodiment has been described above, it is merely an example and does not limit the technical idea of the present specification, and those of ordinary skill in the art to which this specification belongs are within the scope not departing from the essential characteristics of the present embodiment It can be seen that various modifications and applications not exemplified above are possible. That is, each component specifically shown in the embodiment can be implemented by modification. And differences related to such modifications and applications should be construed as being included in the scope of the present specification defined in the appended claims.

Claims

In a method of acquiring high accuracy urination information,

obtaining sound data, the sound data having a start point and an end point;

Acquiring first to mth engraving target data corresponding to m windows from the sound data - Each of the m windows has a predetermined time interval, and continuously between the start point and the end point is determined and consecutive windows among the m windows partially overlap each other, and m is a natural number equal to or greater than 2;

Inputting the first to mth fragment target data into a pre-trained urination/non-urination classification model to obtain first to mth fragment classification data - The urination/non-urination classification model forms a first data matrix learning to receive input and output data including at least one value for classifying a urination section or a non-urination section;

Acquiring first to n-th engraving target data corresponding to n windows from the sound data - The n windows each have a predetermined time interval, and are continuously determined between the start point and the end point. consecutive windows among the n windows partially overlap each other, and n is a natural number of 2 or more and m or less;

Obtaining the first to n-th piece yaw velocity data by inputting the first to n-th piece target data into a pre-trained yaw velocity prediction model - The yaw velocity prediction model receives a second data matrix and at least a value related to the yaw velocity learned to output data containing one or more; and

Containing, including;

How to obtain urination information.
The method of claim 1,

The overlapping degree of successive windows among the m windows is different from the overlapping degree of successive windows among the n windows,

How to obtain urination information.
The method of claim 1,

The overlapping degree of successive windows among the m windows is smaller than the overlapping degree of successive windows among the n windows,

How to obtain urination information.
The method of claim 1,

The first to mth engraving target data and the first to nth engraving target data each include a feature value extracted from at least a portion of the sound data,

How to obtain urination information.
The method of claim 1,

Acquiring the first to mth piece target data comprises:

converting the sound data into spectrogram data; and

Including; obtaining the first to mth slice target data corresponding to the first to mth windows from the spectogram data;

How to obtain urination information.
The method of claim 1,

Acquiring the first to mth piece target data comprises:

acquiring first to mth piece sound data corresponding to the first to mth windows; and

Containing; converting the first to mth sculptural sound data into spectrogram data, respectively, to obtain the first to mth engraving target data;

How to obtain urination information.
7. The method according to claim 5 or 6,

The spectrogram data is Mel-filter applied Mel-spectrogram data,

How to obtain urination information.
The method of claim 1,

The first to mth pieces of classification data are obtained by sequentially inputting the first to mth pieces of target data into the urination/non-urination classification model,

The first to n-th fragment yaw velocity data is obtained by sequentially inputting the first to n-th fragment target data to the yaw velocity prediction model,

How to obtain urination information.
The method of claim 1,

The urination/non-urination classification model comprises at least a first input layer, a first convolution layer, a first hidden layer, and a first output layer, and ,

The yaw velocity prediction model includes at least a second input layer, a second convolutional layer, a second hidden layer, and a second output layer,

How to obtain urination information.
The method of claim 1,

The size of the first data matrix is the same as the size of the second data matrix,

How to obtain urination information.
The method of claim 1,

The step of obtaining the urination data includes:

obtaining urination classification data using the first to mth fragment classification data;

obtaining candidate yaw velocity data using the first to nth piece yaw velocity data; and

Obtaining the urination data by processing the candidate urine velocity data using the urination classification data;

How to obtain urination information.
12. The method of claim 11,

The urination data is obtained by convolution operation of the urination classification data and the candidate urine velocity data,

How to obtain urination information.
The method of claim 1,

Further comprising; determining the presence or absence of a urination process with respect to the sound data using the first to mth fragment classification data obtained after the step of obtaining the first to mth fragment classification data is performed;

If it is determined that a urination process exists with respect to the sound data, obtaining the first to n-th piece target data from the sound data; Acquiring the 1st to nth piece of urine velocity data, and at least the first to mth fragment classification data and the step of obtaining the urination data using the first to nth piece of urine velocity data are performed,

How to obtain urination information.
In a method of acquiring high accuracy urination information,

obtaining sound data by sampling a sound signal obtained by recording a urination process through an external device, the sound data having a start point and an end point;

Acquiring first to n-th engraving target data corresponding to first to n-th windows from the sound data - The first to n-th windows each have a predetermined time interval, the starting point and the determined continuously between the end points, wherein n is a natural number greater than or equal to 2;

Inputting at least a portion of the first to n-th piece target data into a pre-trained urination/non-urination classification model to obtain first to m-th piece classification data - The urination/non-urination classification model is a first receiving a data matrix and learning to output data including at least one value for classifying a urination section or a non-urination section, wherein m is a natural number greater than or equal to 2 and less than or equal to n;

Obtaining the first to n-th piece yaw velocity data by inputting the first to n-th piece target data into a pre-trained yaw velocity prediction model - The yaw velocity prediction model receives a second data matrix and at least a value related to the yaw velocity learned to output data containing one or more; and

Containing, including;

How to obtain urination information.
15. The method of claim 14,

Two consecutive windows of the first to nth windows overlap,

How to obtain urination information.
15. The method of claim 14,

Two consecutive windows among the first to nth windows do not overlap each other,

wherein m is the same as n,

How to obtain urination information.
15. The method of claim 14,

Each of the first to nth pieces of target data includes a feature value extracted from at least a portion of the sound data,

How to obtain urination information.
15. The method of claim 14,

Acquiring the first to n-th piece target data comprises:

converting the sound data into spectrogram data; and

Including; obtaining the first to nth slice target data corresponding to the first to nth windows from the spectogram data;

How to obtain urination information.
15. The method of claim 14,

Acquiring the first to n-th piece target data comprises:

acquiring first to n-th pieces of sound data corresponding to the first to n-th windows; and

Containing the; converting the first to n-th sculptural sound data into spectrogram data, respectively, to obtain the first to n-th engraving target data;

How to obtain urination information.
20. The method of claim 18 or 19,

The spectrogram data is Mel-spectrogram data,

How to obtain urination information.
15. The method of claim 14,

The first to mth piece classification data are obtained by sequentially inputting m pieces of piece target data corresponding to m non-overlapping windows among the first to nth windows to the urination/non-urination classification model,

The first to n-th fragment yaw velocity data is obtained by sequentially inputting the first to n-th fragment target data to the yaw velocity prediction model,

How to obtain urination information.
15. The method of claim 14,

The urination/non-urination classification model comprises at least a first input layer, a first convolution layer, a first hidden layer, and a first output layer, and ,

The yaw velocity prediction model includes at least a second input layer, a second convolution layer, a second hidden layer, and a second output layer,

The data type input to the first input layer and the data type input to the second input layer are the same;

How to obtain urination information.
15. The method of claim 14,

The size of the first data matrix is the same as the size of the second data matrix,

How to obtain urination information.
15. The method of claim 14,

The step of obtaining the urination data includes:

obtaining urination classification data using the first to mth fragment classification data;

obtaining candidate yaw velocity data using the first to nth piece yaw velocity data; and

Obtaining the urination data by processing the candidate urine velocity data using the urination classification data;

How to obtain urination information.
25. The method of claim 24,

The urination data is obtained by convolution operation of the urination classification data and the candidate urine velocity data,

How to obtain urination information.